Dermatology Branch Journal Club

The Cancer Genome Atlas (TCGA)

Assay and analyze cancer samples

HT techniques have allowed researchers to systematically explore the entire spectrum of genomic changes in many cancer types, many individual samples

\(> 200\) forms of cancer, and many more subtypes, caused by errors in DNA that cause cells to grow uncontrolled, there is a lot of space to explore to find oncogenic factors; and, in order to separate the drivers (signal) from passenger (noise) changes, a large number of samples of each tumor type is required.

Hence the goal of the TCGA, which is performing this in more than 20 types of human cancer with sample sizes \(> 1000\) for some cancer types. They are about halfway done…

Identifying the changes in each cancer’s complete set of DNA – its genome – and understanding how such changes interact to drive the disease will lay the foundation for improving cancer prevention, early detection and treatment.

However, that's not the only way this data can be used.

This led to the creation of the Pan-cancer initiative

The TCGA Pan-Cancer Initiative

Focus on TCGA Pan-Cancer Initiative

Nature Genetics October 2013

TCGA analysis

Two fundamental observations

Intra-cancer heterogeneity

Tumors originating in the same tissue vary substantially in genomic alterations.

Cross-cancer similarity

Similar patterns of genomic alteration are observed in tumors from different tissues of origin.

TL;DR

\(3299\) tumors across \(12\) different cancer types
Integrated multiple types of alterations (genomic & epigenomic)
Hierarchical stratification approach to obtain clusters of tumors
Observed two major clusters
- M class: primarily somatic mutation
- C class: primarily copy number alterations
Inverse relationship between # of copy number alterations & # of somatic mutations (when averaged over 12 cancer types)
Oncogenic signatures were used to derive the oncogenic pathways
Nominated therapeutically actionable targets across tumor types

\(3299\) tumors, \(12\) different cancer types

PANCAN12 dataset

Data reduction

Data processing flowchart

Integrated multiple types of alterations

Copy number aberrations (CNAs)

Integrated multiple types of alterations

Copy number aberrations (CNAs)

Integrated multiple types of alterations

Somatic mutations (SNVs)

Integrated multiple types of alterations

Somatic mutations (SNVs)

Integrated multiple types of alterations

DNA methylation

Integrated multiple types of alterations

DNA methylation

Data for further analysis

Selected functional events (SFEs)

treated as binary variables (occured or not occured)
network that connects samples to alterations

Bipartite network analysis

samples connected to alterations

Bipartite network analysis

Find optimal modules

Data for further analysis

Find optimal modules recursively

Hierarchical stratification to obtain clusters

Network modularity

The data can be described as a mathematical structure called a bi-partite graph - nodes are samples or alterations, edges connect a sample that has a specific alteration.

The problem of clustering can then be solved as partitioning the graph into modules based on their connectivity (edges).

This can be applied multiple times within each module to identify sub-modules, etc.

Hence, this method is essentially a type of hierarchical clustering.

For example, we have two node types here that have been partitioned into two modules, and each module has been partitioned into sub-modules.

This network is not one for tumors, but for a dataset to test that their algorithm works from a study of "The Deep South: A Social Anthropological Study of Caste and Class"

Also tested a network of Scottish company relationships in the 20th century…

Hierarchical stratification to obtain clusters of tumors

Observed two major clusters

M(utations) vs C(opy) number alterations

Fig 2a

Features correspond to identified classes

Inverse relationship b/t CNA and SNV

Cancer hyperbol[a|e]

Inverse relationship b/t CNA and SNV

Cancer hyperbol[a|e]

Validation PANCAN18 dataset

Signatures in the M class

each subclass is defined by a core set of alterations common to those tumors (see supplemental figure 5 for details).

The M class included the majority of kidney, glioblast, AML, colo, uterine cancers; these were recursively divided into 17 subclasses.

First subclass had alterations to the PI3K-AKT pathway genes Second subclass had alterations to APC-TP3-KRAS.

Across the M Class alterations included mutations of ARID1A and CTCF, specifically hotspot mutations (including novel hotspot in ARID1A)

In the case of CTCF, variants were found across cancer types (predominantly breast luminbal, endometrial, and a few H&N and kidney) in subclass M5 that hit different zinc fingers, which CTCF uses for sequence selective DNA binding. This indicates selection for zinc finger loss in this subset of samples not that is not tumor type specific

Most M sub-classes were tissue-independent (M1-M14) However, M15-M17 were predominantly tissue-specific subsets (GBM (EGFR), LAML (NPM1), and KIRC (VHL))

Signatures in the C class

Therapeutically actionable targets

Four well studied pathways across tumor types

Therapeutically actionable targets

across tumor types

Conclusions

Two major classes of tumors with different oncogenic drivers (mutation and copy number change)
Argues for tissue-independent classification and treatment of tumors
Almost all solid tumors, many other tumor types not covered
Though the methods should get better with more data

An Analysis is a new analysis of existing data (typically large genomic, transcriptomic or proteomic data sets from arrays or other high-throughput platforms) or new data obtained in a comparative analysis of technologies that lead to novel and arresting conclusions of importance to a broad audience.

Thanks!